1. Field of the Invention
The present invention relates in general to a method and an apparatus for predictively coding a contour of a video signal, and more particularly to a method and an apparatus for comparing a contour (referred to hereinafter as current contour) of the current object image with a contour (referred to hereinafter as previous contour) of the previous object image to obtain a difference therebetween and coding the current contour only when the obtained difference exceeds a predetermined reference value. For example, the present invention is applicable to a shape information coding method of the moving picture experts group-4 (referred to hereinafter as MPEG-4) which is an international standard on moving pictures and audio coding, and other image coding methods considering shape information.
2. Description of the Prior Art
Recently, ISO/IEC/WG11 has considered a method of coding an object with arbitrary shape information, differently from MPEG-1 and MPEG-2 performing frame-unit coding.
Here, a given video is divided into a background image and an object image, and a rectangle including the divided background image and object image is defined as a video object plane (referred to hereinafter as VOP). In MPEG-4, in the case where object regions including desired objects or areas are present in images, they are divided into VOPs and the divided VOPs are coded individually.
Such a VOP has the advantage of freely synthesizing or disintegrating a natural image or an artificial image as the unit of object image. As a result, the VOP is a fundamental factor in processing an object image in fields such as computer graphics, multimedia, etc.
FIG. 1 is a block diagram illustrating the construction of a conventional verification model encoder 10, firstly established by ISO/IEC JTC1/SC29/WG11 MPEG96/N1172 JANUARY. As shown in this drawing, the encoder 10 comprises a VOP formation unit 11, VOP coders 12A, 12B, . . . , 12N and a multiplexer 13.
The VOP formation unit 11 is adapted to input an image sequence to be transmitted or stored, divide the inputted image sequence into object images and form VOPs corresponding respectively to the divided object images.
FIG. 2 is a view illustrating a conventional VOP with shape information, which is partitioned into macro blocks. As shown in this drawing, the VOP contains an object image of "cat". Here, a horizontal size of the VOP is defined as a VOP width and a vertical size thereof is defined as a VOP height. The left top corner of the VOP is defined as a grid start point, and the VOP is partitioned into M.times.N macro blocks, each of which includes M pixels on the X-axis and N pixels on the Y-axis. For example, the VOP may be partitioned into 16.times.16 macro blocks, each of which includes 16 pixels on the X-axis and 16 pixels on the Y-axis.
Noticeably, in the case where macro blocks at the rightmost and bottom portions of the VOP do not include M pixels on the X-axis and N pixels on the Y-axis, respectively, the VOP is enlarged in size in such a manner that the X and Y-axis pixels of each of the macro blocks can be M and N in number, respectively.
Both M and N are set to an even number so that a texture coder can perform sub block-unit coding, as will be mentioned later.
Conventionally, in coding a moving picture in the unit of object, shape information are transmitted together with motion information beginning with that having the highest priority, for the prediction of motion compensation. At this time, different motion information must be applied to adjacent pixels on the object boundary.
Several approaches to representing such an area boundary have been proposed in fields such as computer graphics, character recognition, object synthesis, etc. For example, such approaches may be chain coding, polygon approximation and spline approximation. However, such approaches do not consider transmission. In this connection, it is difficult to transmit coded shape information of a motion region of each frame because of a high transmission rate.
A contour predictive coding method has been suggested to solve the above problem. A high redundancy is present between shape information of a motion region of the same object in successive images. On the basis of such a characteristic, the contour predictive coding method is adapted to perform motion compensation prediction of a contour and transmit the predicted error to reduce a shape information transmission amount. Shape information of a motion region of the same object on successive frames are very analogous in form and position. As a result, the current shape information can be predicted on the basis of the previous shape information. Further, motion information of a moving object is estimated and motion compensation prediction is performed with respect to shape information according to the estimated motion information. In the case where the motion region extraction and motion information estimation are ideally accurate, the transmission of shape information is not necessary.
However, in the above-mentioned contour predictive coding method, the shape information becomes more important as the transmission rate becomes lower. In this connection, an efficient coding method is required to significantly reduce shape information to obtain a higher coding gain than that of a block-unit coding method requiring no transmission of shape information.
In order to solve the problem with the above-mentioned contour predictive coding method, there have been proposed thresholding operation and contour approximation, which are disclosed in U.S. patent application Ser. No. 08/478,558, filed in the name of Hyundai Electronics Industrious Co, Ltd. The thresholding operation and contour approximation will hereinafter be described briefly.
FIG. 5 is a schematic view illustrating shape information of the previous and current frames and FIG. 6 is a schematic view illustrating a variation when the current frame in FIG. 5 is motion-compensated along a motion region of the previous frame in FIG. 5. Nine isolated error areas appear as shown in FIG. 6. The thresholding operation and contour approximation are performed with respect to each of the predictive error areas. A contour of a predictive error area decided on transmission is transmitted according to a characteristic of the corresponding area. At this time, the contour is transmitted approximately rather than accurately for the reduction of shape information. Therefore, information having no effect on the human's eyesight, or information having no effect on the subjective picture quality is not transmitted to make coding at a low transmission rate possible.
FIGS. 7A to 7C are schematic views illustrating polygon and spline approximations. A contour of a predictive error area decided on transmission is first approximated to a polygon. At this time, the number of vertices of the contour is determined according to the polygon approximation level and a spline is obtained along the vertices.
FIGS. 8A to 8C are schematic views illustrating approximation using a predictive contour. First, an error of the predictive contour is obtained for the approximation of a transmission contour of an error area resulting from motion information estimation, not motion region extraction. Then, the transmission contour is drawn on the frame according to the obtained error. In this case, the spline approximation is not necessary.
On the other hand, a contour approximation method, proposed in HOETTER thesis, is to approximate all contours of a motion region, perform motion compensation prediction of vertices used in the approximation and transmit the resultant error. However, such contour approximation method has a disadvantage in that all the contours are always approximated although all of them are not coded.